[X86] Add test case to show failure to combine away negates that may be created by mul by constant expansion.
Mul by constant can expand to a sequence that ends with a negate. If the next instruction is an add or sub we might be able to fold the negate away.
We currently fail to do this because we explicitly don't add anything to the DAG combine worklist when we expand multiplies. This is primarily to keep the multipy from being reformed, but we should consider adding the users to worklist.
[MachineOutliner][NFC] Make Candidates own their call information
Before this, TCI contained all the call information for each Candidate.
This moves that information onto the Candidates. As a result, each Candidate
can now supply how it ought to be called. Thus, Candidates will be able to,
say, call the same function in cheaper ways when possible. This also removes
that information from TCI, since it's no longer used there.
A follow-up patch for the AArch64 outliner will demonstrate this.
[MachineOutliner][NFC] Move missed opt remark into its own function
Having the missed remark code in the middle of `findCandidates` made the
function hard to follow. This yanks that out into a new function,
`emitNotOutliningCheaperRemark`.
[MachineOutliner][NFC] Sink some candidate logic into OutlinedFunction
Just some simple gardening to improve clarity.
Before, we had something along the lines of
1) Create a std::vector of Candidates
2) Create an OutlinedFunction
3) Create a std::vector of pointers to Candidates
4) Copy those over to the OutlinedFunction and the Candidate list
Now, OutlinedFunctions create the Candidate pointers. They're still copied
over to the main list of Candidates, but it makes it a bit clearer what's
going on.
Joel Galenson [Tue, 24 Jul 2018 15:21:54 +0000 (15:21 +0000)]
Use SCEV to avoid inserting some bounds checks.
This patch uses SCEV to avoid inserting some bounds checks when they are not needed. This slightly improves the performance of code compiled with the bounds check sanitizer.
[PredicateInfo] Use custom mangling to support ssa_copy with unnamed types.
This is a workaround and it would be better to fix this generally, but
doing it generally is quite tricky. See D48541 and PR38117.
Doing it in PredicateInfo directly allows us to use the type address to
differentiate different unnamed types, because neither the created
declarations nor the ssa_copy calls should be visible after
PredicateInfo got destroyed.
Simon Atanasyan [Tue, 24 Jul 2018 13:47:52 +0000 (13:47 +0000)]
[mips] Fix local dynamic TLS with Sym64
For the final DTPREL addition, rather than a lui/daddiu/daddu triple,
LLVM was erronously emitting a daddiu/daddiu pair, treating the %dtprel_hi
as if it were a %dtprel_lo, since Mips::Hi expands unshifted for Sym64.
Instead, use a new TlsHi node and, although unnecessary due to the exact
structure of the nodes emitted, use TlsHi for local exec too to prevent
future bugs. Also garbage-collect the unused TprelLo and TlsGd nodes,
and TprelHi since its functionality is provided by the new common TlsHi node.
[x86/SLH] Extract the core register hardening logic to a low-level
helper and restructure the post-load hardening to use this.
This isn't as trivial as I would have liked because the post-load
hardening used a trick that only works for it where it swapped in
a temporary register to the load rather than replacing anything.
However, there is a simple way to do this without that trick that allows
this to easily reuse a friendly API for hardening a value in a register.
That API will in turn be usable in subsequent patcehs.
This also techincally changes the position at which we insert the subreg
extraction for the predicate state, but that never resulted in an actual
instruction and so tests don't change at all.
ADT: Shrink SmallVector size 0 to 16B on 64-bit platforms
SmallVectorTemplateCommon wants to know the address of the first element
so it can detect whether it's in "small size" mode.
The old implementation split the small array, creating the storage for
the first element in SmallVectorTemplateCommon, and pulling the rest
into SmallVectorStorage where we know the size of the array. This
bloats SmallVector size 0 by the larger of sizeof(void*) and sizeof(T),
and we're not even using the storage.
The new implementation leaves the full small storage to
SmallVectorStorage. To calculate the offset of the first element in
SmallVectorTemplateCommon, we just need to know how far to jump, which
we can calculate out-of-band. One subtlety is that we need
SmallVectorStorage to be properly aligned even when the size is 0, to be
sure that (for large alignments) we actually have the padding and it's
well defined to do the pointer math.
[x86] Clean up and convert test to use generated CHECK lines.
This test was already checking microscopic behavior of tail call under
specific conditions. This just makes the CHECK lines much more
consistent, clear, and easily updated when intentional changes are made.
I've also switched the test to consistently name the entry block and to
order the helper declarations and comments for specific tests in the
more usual locations.
We will collect label information from DBG_LABEL. Before every DBG_LABEL,
we will generate a temporary symbol to denote the location of the label.
The symbol could be used to get DW_AT_low_pc afterwards. So, we create a
mapping between 'inlined label' and DBG_LABEL MachineInstr in DebugHandlerBase.
The DBG_LABEL in the mapping is used to query the symbol before it.
The AbstractLabels in DwarfCompileUnit is used to process labels in inlined
functions.
We also keep a mapping between scope and labels in DwarfFile to help to
generate correct tree structure of DIEs.
Tom Stellard [Tue, 24 Jul 2018 01:43:49 +0000 (01:43 +0000)]
AMDGPU/GlobalISel: Remove unnecessary legality constraint for G_EXTRACT
Summary:
We were marking G_EXTRACT operations unsupported if the output type
was larger than the input type. I don't see how this could ever actually
happen, so I dropped the constraint. Doing this makes it possible to
reuse the same legality code for G_INSERT.
Andres Freund [Tue, 24 Jul 2018 00:54:06 +0000 (00:54 +0000)]
Add PerfJITEventListener for perf profiling support.
This new JIT event listener supports generating profiling data for
the linux 'perf' profiling tool, allowing it to generate function and
instruction level profiles.
Currently this functionality is not enabled by default, but must be
enabled with LLVM_USE_PERF=yes. Given that the listener has no
dependencies, it might be sensible to enable by default once the
initial issues have been shaken out.
I followed existing precedent in registering the listener by default
in lli. Should there be a decision to enable this by default on linux,
that should probably be changed.
Please note that until https://reviews.llvm.org/D47343 is resolved,
using this functionality with mcjit rather than orcjit will not
reliably work.
Disregarding the previous comment, here's an example:
$ cat /tmp/expensive_loop.c
bool stupid_isprime(uint64_t num)
{
if (num == 2)
return true;
if (num < 1 || num % 2 == 0)
return false;
for(uint64_t i = 3; i < num / 2; i+= 2) {
if (num % i == 0)
return false;
}
return true;
}
int main(int argc, char **argv)
{
int numprimes = 0;
for (uint64_t num = argc; num < 100000; num++)
{
if (stupid_isprime(num))
numprimes++;
}
[Debugify] Move interface definitions to a header, NFC
This is a minor cleanup in preparation for a change to export DI
statistics from -check-debugify. To do that, it would be cleaner to have
a dedicated header for the debugify interface.
This code was really nasty, had several bugs in it originally, and
wasn't carrying its weight. While on Zen we have all 4 ports available
for SHRX, on all of the Intel parts with Agner's tables, SHRX can only
execute on 2 ports, giving it 1/2 the throughput of OR.
Worse, all too often this pattern required two SHRX instructions in
a chain, hurting the critical path by a lot.
Even if we end up needing to safe/restore EFLAGS, that is no longer so
bad. We pay for a uop to save the flag, but we very likely get fusion
when it is used by forming a test/jCC pair or something similar. In
practice, I don't expect the SHRX to be a significant savings here, so
I'd like to avoid the complex code required. We can always resurrect
this if/when someone has a specific performance issue addressed by it.
Wolfgang Pieb [Mon, 23 Jul 2018 22:37:17 +0000 (22:37 +0000)]
[DWARF v5] Refactor range lists dumping by using a more generic way of handling tables of lists.
The intent is to use it for location list tables as well. Change is almost NFC with the exception
of the spelling of some strings used during dumping (all lowercase now).
Teresa Johnson [Mon, 23 Jul 2018 22:33:57 +0000 (22:33 +0000)]
[LTO] Handle __imp_ (dllimport) symbols consistently with lld
Summary:
Similar to what lld already does for dllimport symbols which are
prefaced with __imp_ (see lld patch r240620), strip off the __imp_
prefix in LTO. Otherwise we can get 2 separate GlobalResolution for
a single symbol, the dllimport declaration, and the definition, which
leads to incorrect LTO handling.
Erik Pilkington [Mon, 23 Jul 2018 22:23:04 +0000 (22:23 +0000)]
[demangler] call terminate() if allocation failed
We really should set *status to memory_alloc_failure, but we need to refactor
the demangler a bit to properly propagate the failure up the stack. Until then,
its better to explicitly terminate then rely on a null dereference crash.
Martin Storsjo [Mon, 23 Jul 2018 22:15:19 +0000 (22:15 +0000)]
[COFF] Fix assembly output of comdat sections without an attached symbol
Since SVN r335286, the .xdata sections are produced without an attached
symbol, which requires using a different syntax when printing assembly
output.
Instead of the usual syntax of '.section <name>,"dr",discard,<symbol>',
use '.section <name>,"dr"' + '.linkonce discard' (which is what GCC
uses for all assembly output).
Martin Storsjo [Mon, 23 Jul 2018 22:15:14 +0000 (22:15 +0000)]
[AArch64] Use MCAsmInfoMicrosoft and MCAsmInfoGNUCOFF as base classes
This matches the structure used on X86 and ARM. This requires
a little bit of duplication of the parts that are equal in both
AArch64 COFF variants though.
Before SVN r335286, these classes didn't add anything that MCAsmInfoCOFF
didn't, but now they do.
This makes AArch64 match X86 in how comdat is used for float constants
for MinGW.
The llvm::Optional data formatter needs to look through the `Storage`
container if it's present.
Before:
220 if (Op && Op->getOp() != dwarf::DW_OP_LLVM_fragment)
-> 221 HasComplexExpression = true;
222
223 // If the register can only be described by a complex expression (i.e.,
224 // multiple subregisters) it doesn't safely compose with another complex
Target 0: (llc) stopped.
(lldb) p Op
(llvm::Optional<llvm::DIExpression::ExprOperand>) $0 = None
After:
(lldb) p Op
(llvm::Optional<llvm::DIExpression::ExprOperand>) $0 =
(llvm::DIExpression::ExprOperand) storage = {
Op = 0x000000010603d460
}
This patch makes debug counters keep track of the total number of times
we've called `shouldExecute` for each counter, so it's easier to build
automated tooling on top of these.
Summary:
Check if the parent basic block and caller exists
before calling CS.getCaller when constant folding
strip.invariant.group instrinsic.
This avoids a crash when the function containing the intrinsic
is being inlined. The instruction is checked for any simplifiction
but has not yet been added to a basic block.
Re-land r335297 "[X86] Implement more of x86-64 large and medium PIC code models"
Don't try to generate large PIC code for non-ELF targets. Neither COFF
nor MachO have relocations for large position independent code, and
users have been using "large PIC" code models to JIT 64-bit code for a
while now. With this change, if they are generating ELF code, their
JITed code will truly be PIC, but if they target MachO or COFF, it will
contain 64-bit immediates that directly reference external symbols. For
a JIT, that's perfectly fine.
David Greene [Mon, 23 Jul 2018 20:23:50 +0000 (20:23 +0000)]
Fix RegScavenger::unprocess
RegScavenger::unprocess walks backward, so it should undo the effects
of defs before undoing effects of kills. Previously it did things in
the opposite order, leaving a register apparently unused (dead) in the
case where an instruction both used (killed) and defined a register.
Paul Semel [Mon, 23 Jul 2018 18:49:04 +0000 (18:49 +0000)]
[yaml2obj] Add default sh_entsize for dynamic sections
Dynamic section holds a table, so the sh_entsize might be set. As the
dynamic section entry size never changes, we can default it to the size
of a dynamic entry.
[lit] Move the shtest-xunit-output check lines into shtest-format
These two tests are operating on the same test suite, which causes
them to be racy about writing temporary files and can cause spurious
failures. Merge them into one test to avoid the issue.
Sam Parker [Mon, 23 Jul 2018 15:25:59 +0000 (15:25 +0000)]
[ARM][NFC] ParallelDSP reorganisation
In preparing to allow ARMParallelDSP pass to parallelise more than
smlads, I've restructed some elements:
- The ParallelMAC struct has been renamed to BinOpChain.
- The BinOpChain struct holds two value lists: LHS and RHS, as well
as inheriting from the OpChain base class.
- The OpChain struct holds all the values of the represented chain
and has had the memory locations functionality inserted into it.
- ParallelMACList becomes OpChainList and it now holds pointers
instead of objects.
Sam Parker [Mon, 23 Jul 2018 12:27:47 +0000 (12:27 +0000)]
[ARM] ARMCodeGenPrepare backend pass
Arm specific codegen prepare is implemented to perform type promotion
on icmp operands, which can enable the removal of uxtb and uxth
(unsigned extend) instructions. This is possible because performing
type promotion before ISel alleviates this duty from the DAG builder
which has to perform legalisation, but has a limited view on data
ranges.
The pass visits any instruction operand of an icmp and creates a
worklist to traverse the use-def tree to determine whether the values
can simply be promoted. Our concern is values in the registers
overflowing the narrow (i8, i16) data range, so instructions marked
with nuw can be promoted easily. For add and sub instructions, we are
able to use the parallel dsp instructions to operate on scalar data
types and avoid overflowing bits. Underflowing adds and subs are also
permitted when the result is only used by an unsigned icmp.
John Brawn [Mon, 23 Jul 2018 12:14:45 +0000 (12:14 +0000)]
[GVN] Don't use the eliminated load as an available value in phi construction
In ConstructSSAForLoadSet if an available value is actually the load that we're
doing SSA construction to eliminate, then we can omit it as SSAUpdate will add
in the value for the phi that will be replacing it anyway. This can result in
simpler IR which can allow further optimisation.
[MemorySSAUpdater] Update Phi operands after trivial Phi elimination
Bug fix for PR37445. The underlying problem and its fix are similar to PR37808.
The bug lies in MemorySSAUpdater::getPreviousDefRecursive(), where PhiOps is
computed before the call to tryRemoveTrivialPhi() and it ends up being out of
date, pointing to stale data. We have now turned each of the PhiOps into a
TrackingVH<MemoryAccess>.
Roman Lebedev [Mon, 23 Jul 2018 10:10:13 +0000 (10:10 +0000)]
[NFC][MCA] ZnVer1: Update RegisterFile to identify false dependencies on partially written registers.
Summary:
Pretty mechanical follow-up for D49196.
As microarchitecture.pdf notes, "20 AMD Ryzen pipeline",
"20.8 Register renaming and out-of-order schedulers":
The integer register file has 168 physical registers of 64 bits each.
The floating point register file has 160 registers of 128 bits each.
"20.14 Partial register access":
The processor always keeps the different parts of an integer register together.
...
An instruction that writes to part of a register will therefore have a false dependence
on any previous write to the same register or any part of it.
Bug fix for PR36787. When reasoning if it's safe to hoist a load we
want to make sure that the defining memory access dominates the new
insertion point of the hoisted instruction. safeToHoistLdSt calls
firstInBB(InsertionPoint,DefiningAccess) which returns false if
InsertionPoint == DefiningAccess, and therefore it falsely thinks
it's safe to hoist.
[x86/SLH] Add a test covering indirect forms of control flow. NFC.
This specifically covers different ways of making indirect calls and
jumps. There are some bugs in SLH that I will be fixing in subsequent
patches where the diff in the generated instructions makes the bug fix
much more clear, so just checking in a baseline of this test to start.
I'm also going to be adding direct mitigation for variant 1.2 which this
file very specifically tests in the various forms it can arise on x86.
Again, the diff to the generated instructions should make the change for
that much more clear, so having the test as a baseline seems useful.
[x86/SLH] Rename and comment the main hardening function. NFC.
This provides an overview of the algorithm used to harden specific
loads. It also brings this our terminology further in line with
hardening rather than checking.
[X86] Remove the max vector width restriction from combineLoopMAddPattern and rely splitOpsAndApply to handle splitting.
This seems to be a net improvement. There's still an issue under avx512f where we have a 512-bit vpaddd, but not vpmaddwd so we end up doing two 256-bit vpmaddwds and inserting the results before a 512-bit vpaddd. It might be better to do two 512-bits paddds with zeros in the upper half. Same number of instructions, but breaks a dependency.
[SelectionDAGBuilder] Use APInt::isZero instead of comparing APInt::getZExtValue to 0 in a place where we can't be sure contents of the APInt fit in a uint64_t.
This is used on an extract vector element index which is most cases is going to be an i32 or i64 and the element will be a valid element number. But it is possible to construct IR with a larger type and large out of range value.
Matt Davis [Sat, 21 Jul 2018 18:32:47 +0000 (18:32 +0000)]
[llvm-mca][docs] Add documentation for the statistic outputs from mca. NFC
Summary: The original text was lifted from the MCA README. I re-ran the dot-product example and updated the output seen in the docs. I also added a few paragraphs discussing the instruction issued and retired histograms, as well as discussing the register file stats.
Simon Atanasyan [Sat, 21 Jul 2018 16:16:03 +0000 (16:16 +0000)]
[mips] Move out the WrapperPat declaration from the NotInMicroMips predicate
This is a follow-up to the rL335185. Those commit adds some WrapperPat
patterns for microMIPS target. But declaration of the WrapperPat class
is under the NotInMicroMips predicate and microMIPS patterns cannot be
selected because predicate (Subtarget->inMicroMipsMode()) &&
(!Subtarget->inMicroMipsMode()) is always false.
This change move out the WrapperPat class declaration from the
NotInMicroMips predicate and enables microMIPS WrapperPat patterns.
If an error occurs and we write it to stderr, it could appear
before we wrote the mangled name which we're undecorating.
By flushing stdout first, we ensure that the messages are always
sequenced in the correct order.
Aaron Smith [Sat, 21 Jul 2018 05:42:13 +0000 (05:42 +0000)]
[DebugInfo] Add a new DI flag to record if a C++ record is a trivial type
Summary:
This flag is used when emitting debug info and is needed to initialize subprogram and member function attributes (function options) for Codeview. These function options are used to create an accurate compiler type for UDT symbols (class/struct/union) from PDBs.
It is not easy to determine if a C++ record is trivial or not based on the current DICompositeType flags and other accessible debug information from Codeview. For example, without this flag the metadata for a non-trivial C++ record with user-defined ctor and a trivial one with a defaulted ctor are the same.
struct S { S(); }
struct S { S() = default; }
This change introduces a new DI flag and corresponding clang::CXXRecordDecl::isTrivial method to set the flag in the frontend.
Change the cap on the amount of padding for each vtable to 32-byte (previously it was 128-byte)
We tested different cap values with a recent commit of Chromium. Our results show that the 32-byte cap yields the smallest binary and all the caps yield similar performance.
Based on the results, we propose to change the cap value to 32-byte.
Roman Tereshin [Fri, 20 Jul 2018 20:10:04 +0000 (20:10 +0000)]
Reapply "[LSV] Refactoring + supporting bitcasts to a type of different size"
This reapplies commit r337489 reverted by r337541
Additionally, this commit contains a speculative fix to the issue reported in r337541
(the report does not contain an actionable reproducer, just a stack trace)
Joel E. Denny [Fri, 20 Jul 2018 20:09:56 +0000 (20:09 +0000)]
[FileCheck] Fix search ranges for DAG-NOT-DAG
A DAG-NOT-DAG is a CHECK-DAG group, X, followed by a CHECK-NOT group,
N, followed by a CHECK-DAG group, Y. Let y be the initial directive
of Y. This patch makes the following changes to the behavior:
1. Directives in N can no longer match within part of Y's match
range just because y happens not to be the earliest match from
Y. Specifically, this patch withdraws N's search range end
from y's match range start to Y's match range start.
2. y can no longer match within X's match range, where a y match
produced a reordering complaint, which is thus no longer
possible. Specifically, this patch withdraws y's search range
start from X's permitted range start to X's match range end,
which was already the search range start for other members of
Y.
Both of these changes can only increase the number of test passes: #1
constrains the ability of CHECK-NOTs to match, and #2 expands the
ability of CHECK-DAGs to match without complaints.
1. These changes simplify the FileCheck conceptual model. First,
it makes search ranges for DAG-NOT-DAG more consistent with
other cases. Second, it was confusing that y was treated
differently from the rest of Y.
2. These changes add theoretical use cases for DAG-NOT-DAG that
had no obvious means to be expressed otherwise. We can justify
the first half of this assertion with the observation that
these changes can only increase the number of test passes.
3. Reordering detection for DAG-NOT-DAG had no obvious real
benefit.
We don't have evidence from real uses cases to help us debate
conclusions #2 and #3, but #1 at least seems intuitive.
Jordan Rupprecht [Fri, 20 Jul 2018 19:54:24 +0000 (19:54 +0000)]
[llvm-objcopy] Add basic support for --rename-section
Summary:
Add basic support for --rename-section=old=new to llvm-objcopy.
A full replacement for GNU objcopy requires also modifying flags (i.e. --rename-section=old=new,flag1,flag2); I'd like to keep that in a separate change to keep this simple.
projects / llvm / log